Learn · Article
Your Bot Is Good at the Middle. Give It Both Ends.
Most jobs slip at the start, when nobody notices them, or at the end, when nobody closes them out. Let your helper find the work and finish it, and keep an hour to say no.
Most people think the hard part of a job is doing it. Writing the reply. Building the quote. Filling out the form.
Usually it isn't. The hard part is the two ends. Somebody has to notice the job exists, and somebody has to close it out. In most small shops and most households, that somebody is you.
A new lead comes in through the website on Tuesday while you're under a truck. You see it Thursday night. Writing the reply takes four minutes. Finding it took two days. Then it still has to go out, and somebody has to check back next week.
That four-minute middle is where most people use their bot. It's also the part they needed help with least.
Lauren (@poteto) put a name on this. Lauren's favorite use for Grok Bot is "the first and the last mile of getting work done." The first mile is figuring out what work you should even be doing. The last mile is following through and closing the loop.
We've talked before about how work dies between bots when nobody owns the next move. The same thing happens at the two ends of a job. Usually the one dropping it is the person who was supposed to notice.
The first mile: let it find the work
If your helper waits for you to tell it what to do, you're still doing the noticing.
The first mile starts when you connect it to the places where jobs already show up: the inbox, the calendar, the shared folder where quotes live. Lauren's list adds a customer database and a team chat. Then, instead of spelling out every task, you point it at those places and tell it what to look for. The point, in Lauren's words, is to give it context instead of micromanaging it.
Then you put that look on a schedule. On our own desk, one helper sweeps the shop inbox and messages about every hour and a half through the day. It sorts what came in and drafts the replies. Lauren points out that routines can also kick off when something new shows up.
In a small shop, that sounds like: "Every morning, look at yesterday's emails and messages. Tell me which ones are new customers, which ones are waiting on me, and which ones can wait." At home it sounds like: "Watch the school emails and the family calendar. Tell me what needs a signature, a payment, or a ride this week."
You stop hunting for the work. The work gets brought to you.
The last mile: carry it to done
The other end is where good intentions die. The reply got drafted and never sent. The estimate got written and never followed up. The grocery list got made and never turned into an order.
The last mile means the helper doesn't stop at "here's a draft." It carries the job as far as you've said it can: send the reply, put the follow-up on the calendar, check back in a week, and tell you when it's closed.
I designed our dinner helper around that line. It plans the week, builds the list, and can stage the cart. Then a person checks out. It carries the job right up to the money, and stops there on purpose.
The hour you keep
Here's the part of Lauren's setup worth stealing.
That daily loop is a software one. It watches a team chat for feedback about an app, makes a ticket, has coding agents reproduce and fix the problem, checks the fix, and sends a ping. Then the last step finishes on its own after one hour, unless Lauren asks for changes.
That's not a button you flip. It's a rule written into the loop, and you can write it into yours in plain words: "When it's ready, show me exactly what you're about to do. If I don't say stop within an hour, go ahead."
That hour is the safety valve. You see the reply before the customer does. You can change a word, or kill it. And if you're busy, the work doesn't sit waiting for you to push a button.
Start that rule only on things that can be undone or cost little if they're off: a reminder text, a calendar hold, a "got your message, we'll call tomorrow." Money doesn't go on a timer. Neither does anything you can't take back. Anything sent under your name stays your call; the window just lets you keep it without being the bottleneck. On our desk, anything going out under my name still waits for a plain yes.
One more habit helps here. If you already write down every time a helper parks or refuses something, add one more line: stopped by me, and why. After a month, those lines tell you exactly what to fix in the loop.
Build one loop
Don't automate the whole shop. Pick one job that keeps dying at one end.
A service shop might pick new leads:
1. Check the website form and the inbox every morning.
2. Pull out anything from a new customer.
3. Draft a short reply and put a follow-up on the calendar for three days later.
4. Show me the reply. If I don't say stop within an hour, send it.
5. Three days later, if they haven't answered, tell me.
A household might pick school paperwork: find it, list what's due, put it on the calendar, remind whoever's on the hook the night before. The signature is still yours.
Run it for a couple of weeks. Watch what you stop, and why.
What it won't fix
Not every job that falls through is a bot problem. If the thing that keeps dying at the end is a hard conversation with a customer or a job that needs someone on site, a helper won't close it for you. Sometimes the answer is a specialist on the ground, not a bot.
But a lot of what slips is just noticing and finishing, and that part you can hand off.
Let it notice. Let it finish. Keep the hour.
Do this for me
Copy this into your Grok Bot chat, then fill in your own details.
Every morning, look at yesterday's emails and messages. Tell me which ones are new customers, which ones are waiting on me, and which ones can wait.